May 11, 2026
Git a grip: job dead or drama alive?
From Today, Software Engineering Is Dead
Coders declare the job "dead" and the comments instantly turn into a food fight
TLDR: The article argues AI can now write a lot of code, but human judgment still decides whether that code is any good. Commenters were far less calm, with some saying this proves many coding jobs were never that skilled, while others fear a flood of low-quality AI junk will make real knowledge harder to find.
A spicy new post from Building #138 basically says software engineering isn’t gone, but it’s being radically reshaped by AI code-writing tools. The writer compares it to photography changing painting: the machine can now crank out workable code fast, but humans still matter for judgment, taste, and deciding what should be built in the first place. In other words, the boring part may be getting automated, but the smart part isn’t packing its bags just yet.
But honestly? The comments are where the real fireworks are. One camp flat-out refused the premise, saying if an AI can “slop it out,” then maybe that task never needed a true engineer to begin with. Ouch. Another commenter went full doom mode, warning that AI-generated junk could bury the actually useful guides and knowledge people rely on today, leaving future developers to basically dig civilization out of an algorithmic landfill. That’s not a comment, that’s a disaster movie pitch.
Then came the personal jabs. One reader side-eyed the author’s claim that big design changes used to take weeks or months, bluntly suggesting that if that’s normal, maybe the problem isn’t the work — it’s the worker. Meanwhile others laughed at the idea that using AI gives you more control than writing the code yourself, which is the nerd equivalent of saying a microwave gives you more control than cooking. The mood? Equal parts panic, eye-rolls, and “nice try, but we’ve seen this movie before.”
Key Points
- •The article argues that LLM coding agents are changing software engineering in a way analogous to how photography changed painting rather than eliminating it.
- •The author says tools such as Claude Code now produce more usable code than they did a year earlier, with fewer output issues that break execution.
- •The article states that LLMs reduce the cost and time required to implement or refactor design decisions, making experimentation easier.
- •The article says unguided LLM-generated code can include unnecessary features, architectural assumptions, and excessively large functions.
- •The conclusion is that skilled human oversight remains necessary to evaluate design quality and guide LLM output, even as code generation becomes cheaper and faster.